Observability, Distributed Tracing & the Complex World
summary
Claims that distributed request tracing is the practical trace model (every request needs a unique ID) and that traces are high-volume, high-cardinality so storage and write throughput matter. Covers span/spanID/traceID semantics, B3 propagation, Zipkin and Jaeger exporters, Prometheus and Istio/service mesh integration, and a Google Hipster Store demo that shows instrumentation. Notes logs remain the source of truth for root cause, traces speed causal ordering, and cautions that traces can reach 200 spans per request—driving the need for the OpenTracing/OpenCensus merger.
AI-written summary · about
chapters (27)
- 0:00Introduction
- 1:55Performance Monitoring
- 2:20Distributed Systems
- 2:55Observability
- 4:15Observability is SignaltoNoise
- 5:15Distributed Tracing is Wrong
- 6:05Tracing Analogs
- 7:00Bringing Events Together
- 9:40Span Elements
- 11:00Service Elements
- 11:55Tracers
- 13:25Propagation
- 14:20Standards
- 15:50Open Tracing
- 16:10Open Tracing Limitations
- 17:55Open Tracing API
- 19:15Open Census
- 20:40Open Sensing
- 22:20Viewpoint
- 23:15Languages
- 25:05Logging
- 26:55Open Tracing Open Census merger
- 28:25Hipster Store Demo
- 30:00GitHub Repository
- 30:50How to get involved
- 32:00Can I store traces everywhere
- 34:25I already have a log
related talks
34mReduce Complexity with Distributing Tracing
49mDistributed Tracing: How the Pros Debug Concurrent and Distributed Systems
59mAn introduction to distributed tracing
57mDistributed Tracing: How the Pros Debug Concurrent and Distributed Systems
16mThe State of Observability
16mPragmatic execution tracing
25mTracing Go
23mIntroduction to OpenTracing